Add supervisor.get_previous_traceback() function.

[hierophect]

New PR to avoid force pushing rebase over the old one. In retrospect, I don't think this needs any special interaction with deep sleep, since it's really for passing tracebacks and deep sleep exceptions don't have those. If code needs to detect if the last exception was deepsleep vs startup, it should just use the alarm module as per usual. Tested on a Feather M4 Express, works as expected. Useful for #1084.

Heads up for @tannewt to take a look at the string stuff in this since I'm not sure I'm qualified to comment on whether it's safe. Also to double check that this doesn't have any negative interactions with the new structure of status LED - we don't have any exception specific colors anymore, right?

[tannewt]

Heads up for @tannewt to take a look at the string stuff in this since I'm not sure I'm qualified to comment on whether it's safe.

I'll do my best. :-)

Also to double check that this doesn't have any negative interactions with the new structure of status LED - we don't have any exception specific colors anymore, right?

Correct. Just double red flash.

[tannewt]

Looks good to me! Thank you @cwalther and @hierophect

[cwalther]

Thank you @hierophect!

@tannewt, if the string stuff looks good to you, I guess the TODO comments can be removed? I can make a PR to that effect.

diff --git a/shared-bindings/supervisor/__init__.c b/shared-bindings/supervisor/__init__.c
index 83b473616..055b3a908 100644
--- a/shared-bindings/supervisor/__init__.c
+++ b/shared-bindings/supervisor/__init__.c
@@ -267,14 +267,13 @@ MP_DEFINE_CONST_FUN_OBJ_0(supervisor_ticks_ms_obj, supervisor_ticks_ms);
 //|     ...
 //|
 STATIC mp_obj_t supervisor_get_previous_traceback(void) {
-    // TODO is this a safe and proper way of making a heap-allocated string object that points at long-lived (and likely long and unique) data outside the heap?
     if (prev_traceback_allocation) {
         size_t len = strlen((const char *)prev_traceback_allocation->ptr);
         if (len > 0) {
             mp_obj_str_t *o = m_new_obj(mp_obj_str_t);
             o->base.type = &mp_type_str;
             o->len = len;
-            // TODO is it a good assumption that callers probably aren't going to compare this string, so skip computing the hash?
+            // callers probably aren't going to compare this string, so skip computing the hash
             o->hash = 0;
             o->data = (const byte *)prev_traceback_allocation->ptr;
             return MP_OBJ_FROM_PTR(o);

Also, no need to make it optional anymore? All the better!

[tannewt]

Yes, please make a PR @cwalther. Thanks!